ChanServ changed the topic of #zink to: official development channel for the mesa3d zink driver || https://docs.mesa3d.org/drivers/zink.html
<fdobridge> <b​ornecrantz> Does this use timeline semaphores both ways, or only one way?
<fdobridge> <g​fxstrand> No timeline semaphores except to trigger a submit thread. I don't think any Mesa drivers support the NV extension for timeline semaphores.
<fdobridge> <g​fxstrand> And I wouldn't trust any to do it except Zink.
<fdobridge> <g​fxstrand> And I only "trust" Zink because it calls into Vulkan.
<fdobridge> <b​ornecrantz> Ah okay, thanks for the info.
<fdobridge> <b​ornecrantz> Yeah last I checked there weren't a way to share a timeline semaphore between Vulkan and GL, wish it were tho so I didn't have to use fences.
<fdobridge> <b​ornecrantz> Ah okay, thanks for the info.
<fdobridge> <b​ornecrantz> Yeah last I checked there weren't a way to share a timeline semaphore between Vulkan and GL, wish it were tho so I didn't have to use fences. Or well, a way to import a timeline semaphore. (edited)
<fdobridge> <b​ornecrantz> Ah okay, thanks for the info.
<fdobridge> <b​ornecrantz> Yeah last I checked there weren't a way to share a timeline semaphore between Vulkan and GL, wish it were tho so I didn't have to use fences. Or well, a way to import a timeline semaphore into GL. (edited)
<fdobridge> <g​fxstrand> There's an NV extension, I think. But I would probably NAK any Mesa GL implementation except Zink on the grounds that it's just too darn hard to get right. :frog_sweat:
karolherbst has quit [Ping timeout: 480 seconds]
Sid127- has joined #zink
Sid127 has quit [Ping timeout: 480 seconds]
karolherbst has joined #zink
<fdobridge> <g​fxstrand> IDK if this is a regression or not but chrome doesn't even start with Zink+ANV anymore
<fdobridge> <g​fxstrand> Spamming
<fdobridge> <g​fxstrand> ```
<fdobridge> <g​fxstrand> [2207532:2207532:0224/100059.943961:ERROR:gpu_process_host.cc(953)] GPU process exited unexpectedly: exit_code=134
<fdobridge> <g​fxstrand> MESA: warning: INTEL_HWCONFIG_MIN_HS_URB_ENTRIES (0) != devinfo->urb.min_entries[MESA_SHADER_TESS_CTRL] (1)
<fdobridge> <g​fxstrand> chrome: ../src/gallium/drivers/zink/zink_resource.c:1747: add_resource_bind: Assertion `mod == DRM_FORMAT_MOD_INVALID || new_obj->modifier == DRM_FORMAT_MOD_LINEAR' failed.
<fdobridge> <g​fxstrand> ```
<fdobridge> <z​mike.> :grimace:
<fdobridge> <z​mike.> what is the modifier ?
<fdobridge> <g​fxstrand> Ugh... Let me look.
<fdobridge> <g​fxstrand> Ugh. It's chrome so you can't launch it with a debugger. 😭
<zmike> you can
<zmike> --in-process-gpu
<zmike> unless that's been removed recently
<fdobridge> <g​fxstrand> ```
<fdobridge> <g​fxstrand> (gdb) p mod
<fdobridge> <g​fxstrand> $3 = 0
<fdobridge> <g​fxstrand> (gdb) p new_obj->modifier
<fdobridge> <g​fxstrand> $4 = 72057594037927935
<fdobridge> <g​fxstrand> (gdb) p/x new_obj->modifier
<fdobridge> <g​fxstrand> $5 = 0xffffffffffffff
<fdobridge> <g​fxstrand> (gdb)
<fdobridge> <g​fxstrand> ```
<fdobridge> <z​mike.> yeah so that's invalid
<fdobridge> <z​mike.> uhh
<fdobridge> <z​mike.> I've been tightening up dmabuf exports because previously it was possible to do some weird stuff where it would try to export, fall back to linear tiling (not even LINEAR modifier), and then try to carry on normally
<fdobridge> <z​mike.> and there's somehow no vvl error here even though I'm pretty sure it's not intended
<fdobridge> <g​fxstrand> Still seems to work on NVK
<fdobridge> <z​mike.> yeah I think I missed a conditional here
<fdobridge> <z​mike.> I assume `res->linear` is true?
eukara has joined #zink